Improvement and International Validation of the Predictive Probability of the Patient Demographics, Radiographic Index, and Surgical Invasiveness for Mechanical Failure (PRISM) Model for Preventive Procedures in Adult Spinal Deformity Surgery

Study Design.This is an international multicenter retrospective review of 219 surgically treated consecutive adult spinal deformity (ASD) patients who had a minimum of five fused segments, completed a 2-year follow-up.Objective.The purpose of this study was to add the indices of preventive procedures to improve and to validate the predictive probability of the PRISM (patient demographics, radiographic index, and surgical invasiveness for mechanical failure) for mechanical failure (MF) following ASD surgery.Summary of Background Data.The PRISM was developed from the data of 321 ASD patients, which stratified the risk of MF from six types of risk.Methods.Data from 136 Japanese ASD patients (age 49 ± 21 yr, 88% female) were used to develop PRISM2, and data from 83 US ASD patients (age 58 ± 12 yr, 86% female) were used for the external validation. We analyzed the associations between three preventive procedures (UIV+1 tethering [TH], teriparatide [TP], and multirod [MR]) and MF by multivariate logistic regression analysis (MRA). The values for the nearest integer of the β of the procedures were added to the six indices of the original PRISM to establish the PRISM2. The discriminative ability of the PRISM/ PRISM2 for MF was evaluated using the area under the receiver operating characteristic curve (AUC) and the precision-recall (PR) curve. The Cochran-Armitage test was used to analyze the trend between PRISM/PRISM2 scores and MF.Results.MF developed in 25% (34 cases). The β values for the preventive procedures calculated by MRA were TH: -2.5, TP: -3.0, and MR: -2.1. The Cochran-Armitage test showed an excellent trend between MF and PRISM/2. The diagnostic ability was superior for the PRISM2 compared with the PRISM (PRISM2; AUC = 0.94 [0.90-0.98], PRISM; AUC = 0.87 [0.81-0.93], difference = -0.07 [-0.11 to -0.03], P < 0.01). The AUC of the PRISM2 was 0.70 [0.59-0.81, P < 0.01] in the US patient cohort.Conclusion.We refined the PRISM by adding preventive procedures to the risk indices. Further validation and adjustment in a large different patient cohorts may improve the predictive probability of PRISM2.Level of Evidence: 3.